Skip to main content
SearchLoginLogin or Signup

Moving Speckle Imaging into New Frontiers

Presentation #302.05 in the session Computation, Data Handling, Image Analysis — iPoster Session.

Published onJun 29, 2022
Moving Speckle Imaging into New Frontiers

A common method used in speckle interferometric analysis is based on a series of temporal correlations between Fourier components of short exposure images. We present results on the next advancement beyond speckle interferometry that uses multi-frame blind deconvolution (MFBD) algorithm to detect closely spaced objects with high-contrast ratios, such as faint binary companions. MFBD estimates the parameters that describe the object and the point-spread functions (PSFs) and uses physical constraints to increase the fidelity of these parameter estimates. However, detecting faint companions requires large volumes of data, typically thousands of frames. Numerical algorithms such as MFBD require minimizing an error metric between the modeled data and actual imagery which requires solving a set of parameters that describe the blur in the image and the object scene. In large data sets, parameter space becomes heavily pocketed with local minima, which can typically cause MFBD algorithms to stagnate and fail to find reasonable approximate physical solutions that describe faint companions within the image. We present a compact MFBD (CMFBD) method as a preliminary step before using MFBD that uses consistency constraints imposed on the data from turbulence-induced temporal correlations to move the parameter space closer to the global minimum. We show images of various examples of restorations of objects and compare our results to those produced from conventional speckle interferometric methods. Even though CMFBD/MFBD is at the frontier of speckle image restoration, we plan to improve upon the MFBD algorithm in the near future by introducing the use of alternating direction method of multipliers (ADMM) to improve object and PSF estimation by enforcing physical constraints on sparsity and smoothness in the object, and wavelength diversity as a constraint of the estimated PSFs. ADMM techniques can cascade penalty functions and can leverage on the PSF in multiple ways, simultaneously, for better recovery of the true object.

No comments here